from sklearn_benchmarks.reporting.hp_match import HPMatchReporting

scikit-learn vs. ONNX Runtime

reporting = HPMatchReporting(against_lib="onnx", config="config.yml")
reporting.make_report()

We assume here there is a perfect match between the hyperparameters of both librairies. For a given set of parameters and a given dataset, we compute the speedup time scikit-learn / time onnx. For instance, a speedup of 2 means that onnx is twice as fast as scikit-learn for a given set of parameters and a given dataset.

KNeighborsClassifier_brute_force

onnx (1.10.1) vs. scikit-learn (1.0.dev0)

Speedup barplots

All estimators share the following parameters: algorithm=brute.

WARNING! Mismatch between validation scores for 12 predictions (75.0%). See below.

	estimator	function	n_samples_train	n_samples	n_features	accuracy_score_sklearn	accuracy_score_onnx	diff_accuracy_scores
1	KNeighborsClassifier_brute_force	predict	100000	1000	100	0.663	1.000	0.337
2	KNeighborsClassifier_brute_force	predict	100000	1	100	1.000	0.757	0.243
4	KNeighborsClassifier_brute_force	predict	100000	1000	100	0.757	0.882	0.125
7	KNeighborsClassifier_brute_force	predict	100000	1000	100	0.882	1.000	0.118
8	KNeighborsClassifier_brute_force	predict	100000	1	100	1.000	0.757	0.243
10	KNeighborsClassifier_brute_force	predict	100000	1000	100	0.882	0.663	0.219
13	KNeighborsClassifier_brute_force	predict	100000	1000	100	0.757	1.000	0.243
14	KNeighborsClassifier_brute_force	predict	100000	1	100	1.000	0.922	0.078
16	KNeighborsClassifier_brute_force	predict	100000	1000	100	0.663	0.929	0.266
19	KNeighborsClassifier_brute_force	predict	100000	1000	2	0.896	1.000	0.104
20	KNeighborsClassifier_brute_force	predict	100000	1	2	1.000	0.922	0.078
22	KNeighborsClassifier_brute_force	predict	100000	1000	2	0.922	0.896	0.026

Raw results

	estimator	function	n_samples_train	n_samples	n_features	mean_duration_sklearn	std_duration_sklearn	iteration_throughput	latency	n_jobs	n_neighbors	mean_duration_onnx	std_duration_onnx	accuracy_score_onnx	speedup	std_speedup	sklearn_profiling	onnx_profiling
0	KNeighborsClassifier_brute_force	fit	100000	100000	100	0.011	0.0	7.144	0.0	-1	1	17.218	0.127	0.663	0.001	0.001	See	See
3	KNeighborsClassifier_brute_force	fit	100000	100000	100	0.011	0.0	7.103	0.0	-1	5	0.314	0.007	1.000	0.036	0.036	See	See
6	KNeighborsClassifier_brute_force	fit	100000	100000	100	0.011	0.0	7.121	0.0	1	100	17.078	0.006	0.882	0.001	0.001	See	See
9	KNeighborsClassifier_brute_force	fit	100000	100000	100	0.011	0.0	7.080	0.0	-1	100	0.307	0.007	1.000	0.037	0.037	See	See
12	KNeighborsClassifier_brute_force	fit	100000	100000	100	0.011	0.0	7.218	0.0	1	5	3.767	0.007	0.896	0.003	0.003	See	See
15	KNeighborsClassifier_brute_force	fit	100000	100000	100	0.011	0.0	7.196	0.0	1	1	0.238	0.004	1.000	0.047	0.047	See	See
18	KNeighborsClassifier_brute_force	fit	100000	100000	2	0.004	0.0	0.381	0.0	-1	1	3.700	0.031	0.929	0.001	0.001	See	See
21	KNeighborsClassifier_brute_force	fit	100000	100000	2	0.004	0.0	0.380	0.0	-1	5	0.238	0.004	1.000	0.018	0.018	See	See

	estimator	function	n_samples_train	n_samples	n_features	mean_duration_sklearn	std_duration_sklearn	iteration_throughput	latency	n_jobs	n_neighbors	mean_duration_onnx	std_duration_onnx	accuracy_score_onnx	speedup	std_speedup	sklearn_profiling	onnx_profiling
1	KNeighborsClassifier_brute_force	predict	100000	1000	100	2.033	0.165	0.000	0.002	-1	1	0.311	0.007	1.000	6.528	6.530	See	See
2	KNeighborsClassifier_brute_force	predict	100000	1	100	0.024	0.004	0.000	0.024	-1	1	17.034	0.014	0.757	0.001	0.001	See	See
4	KNeighborsClassifier_brute_force	predict	100000	1000	100	2.877	0.023	0.000	0.003	-1	5	17.121	0.025	0.882	0.168	0.168	See	See
5	KNeighborsClassifier_brute_force	predict	100000	1	100	0.023	0.002	0.000	0.023	-1	5	0.317	0.005	1.000	0.074	0.074	See	See
7	KNeighborsClassifier_brute_force	predict	100000	1000	100	2.047	0.004	0.000	0.002	1	100	0.306	0.008	1.000	6.701	6.703	See	See
8	KNeighborsClassifier_brute_force	predict	100000	1	100	0.020	0.000	0.000	0.020	1	100	16.894	0.045	0.757	0.001	0.001	See	See
10	KNeighborsClassifier_brute_force	predict	100000	1000	100	2.857	0.019	0.000	0.003	-1	100	17.046	0.139	0.663	0.168	0.168	See	See
11	KNeighborsClassifier_brute_force	predict	100000	1	100	0.024	0.002	0.000	0.024	-1	100	0.317	0.006	1.000	0.075	0.075	See	See
13	KNeighborsClassifier_brute_force	predict	100000	1000	100	2.017	0.003	0.000	0.002	1	5	0.239	0.006	1.000	8.432	8.434	See	See
14	KNeighborsClassifier_brute_force	predict	100000	1	100	0.020	0.000	0.000	0.020	1	5	3.725	0.021	0.922	0.005	0.005	See	See
16	KNeighborsClassifier_brute_force	predict	100000	1000	100	1.152	0.002	0.001	0.001	1	1	3.773	0.021	0.929	0.305	0.305	See	See
17	KNeighborsClassifier_brute_force	predict	100000	1	100	0.019	0.000	0.000	0.019	1	1	0.240	0.004	1.000	0.078	0.078	See	See
19	KNeighborsClassifier_brute_force	predict	100000	1000	2	1.816	0.032	0.000	0.002	-1	1	0.237	0.005	1.000	7.646	7.647	See	See
20	KNeighborsClassifier_brute_force	predict	100000	1	2	0.006	0.005	0.000	0.006	-1	1	3.725	0.009	0.922	0.002	0.002	See	See
22	KNeighborsClassifier_brute_force	predict	100000	1000	2	2.721	0.030	0.000	0.003	-1	5	3.767	0.010	0.896	0.722	0.722	See	See
23	KNeighborsClassifier_brute_force	predict	100000	1	2	0.010	0.005	0.000	0.010	-1	5	0.236	0.004	1.000	0.042	0.042	See	See

KNeighborsClassifier_kd_tree

onnx (1.10.1) vs. scikit-learn (1.0.dev0)

Speedup barplots

All estimators share the following parameters: algorithm=kd_tree.

WARNING! Mismatch between validation scores for 12 predictions (75.0%). See below.

	estimator	function	n_samples_train	n_samples	n_features	accuracy_score_sklearn	accuracy_score_onnx	diff_accuracy_scores
1	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	0.929	1.000	0.071
2	KNeighborsClassifier_kd_tree	predict	1000000	1	10	1.000	0.946	0.054
4	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	0.946	0.951	0.005
7	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	0.951	1.000	0.049
8	KNeighborsClassifier_kd_tree	predict	1000000	1	10	1.000	0.946	0.054
10	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	0.951	0.929	0.022
13	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	0.946	1.000	0.054
14	KNeighborsClassifier_kd_tree	predict	1000000	1	10	1.000	0.911	0.089
16	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	0.929	0.894	0.035
19	KNeighborsClassifier_kd_tree	predict	1000	1000	2	0.891	1.000	0.109
20	KNeighborsClassifier_kd_tree	predict	1000	1	2	1.000	0.911	0.089
22	KNeighborsClassifier_kd_tree	predict	1000	1000	2	0.911	0.891	0.020

Raw results

	estimator	function	n_samples_train	n_samples	n_features	mean_duration_sklearn	std_duration_sklearn	iteration_throughput	latency	n_jobs	n_neighbors	mean_duration_onnx	std_duration_onnx	accuracy_score_onnx	speedup	std_speedup	sklearn_profiling	onnx_profiling
0	KNeighborsClassifier_kd_tree	fit	1000000	1000000	10	2.921	0.092	0.027	0.0	-1	1	109.598	0.000	0.929	0.027	0.027	See	See
3	KNeighborsClassifier_kd_tree	fit	1000000	1000000	10	3.562	0.032	0.022	0.0	-1	5	2.599	0.161	1.000	1.370	1.373	See	See
6	KNeighborsClassifier_kd_tree	fit	1000000	1000000	10	3.640	0.059	0.022	0.0	1	100	107.229	0.000	0.951	0.034	0.034	See	See
9	KNeighborsClassifier_kd_tree	fit	1000000	1000000	10	3.566	0.052	0.022	0.0	-1	100	2.583	0.137	1.000	1.381	1.382	See	See
12	KNeighborsClassifier_kd_tree	fit	1000000	1000000	10	3.516	0.080	0.023	0.0	1	5	0.047	0.017	0.891	75.383	80.150	See	See
15	KNeighborsClassifier_kd_tree	fit	1000000	1000000	10	3.571	0.037	0.022	0.0	1	1	0.005	0.000	1.000	709.421	709.599	See	See
18	KNeighborsClassifier_kd_tree	fit	1000	1000	2	0.001	0.001	0.023	0.0	-1	1	0.060	0.001	0.894	0.012	0.012	See	See
21	KNeighborsClassifier_kd_tree	fit	1000	1000	2	0.001	0.000	0.029	0.0	-1	5	0.005	0.000	1.000	0.110	0.111	See	See

	estimator	function	n_samples_train	n_samples	n_features	mean_duration_sklearn	std_duration_sklearn	iteration_throughput	latency	n_jobs	n_neighbors	mean_duration_onnx	std_duration_onnx	accuracy_score_onnx	speedup	std_speedup	sklearn_profiling	onnx_profiling
1	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	0.827	0.991	0.000	0.001	-1	1	2.598	0.183	1.000	0.318	0.319	See	See
2	KNeighborsClassifier_kd_tree	predict	1000000	1	10	0.003	0.000	0.000	0.003	-1	1	107.842	0.000	0.946	0.000	0.000	See	See
4	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	1.027	0.278	0.000	0.001	-1	5	108.026	0.000	0.951	0.010	0.010	See	See
5	KNeighborsClassifier_kd_tree	predict	1000000	1	10	0.003	0.001	0.000	0.003	-1	5	2.599	0.155	1.000	0.001	0.001	See	See
7	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	5.487	0.359	0.000	0.005	1	100	2.590	0.147	1.000	2.118	2.122	See	See
8	KNeighborsClassifier_kd_tree	predict	1000000	1	10	0.003	0.001	0.000	0.003	1	100	107.248	0.000	0.946	0.000	0.000	See	See
10	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	3.044	0.124	0.000	0.003	-1	100	108.632	0.000	0.929	0.028	0.028	See	See
11	KNeighborsClassifier_kd_tree	predict	1000000	1	10	0.005	0.001	0.000	0.005	-1	100	2.580	0.143	1.000	0.002	0.002	See	See
13	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	1.666	0.236	0.000	0.002	1	5	0.005	0.000	1.000	332.285	332.354	See	See
14	KNeighborsClassifier_kd_tree	predict	1000000	1	10	0.002	0.000	0.000	0.002	1	5	0.041	0.009	0.911	0.037	0.038	See	See
16	KNeighborsClassifier_kd_tree	predict	1000000	1000	10	0.893	0.181	0.000	0.001	1	1	0.060	0.003	0.894	14.936	14.954	See	See
17	KNeighborsClassifier_kd_tree	predict	1000000	1	10	0.001	0.000	0.000	0.001	1	1	0.005	0.000	1.000	0.225	0.225	See	See
19	KNeighborsClassifier_kd_tree	predict	1000	1000	2	0.025	0.014	0.001	0.000	-1	1	0.005	0.000	1.000	4.983	4.984	See	See
20	KNeighborsClassifier_kd_tree	predict	1000	1	2	0.002	0.000	0.000	0.002	-1	1	0.038	0.001	0.911	0.052	0.052	See	See
22	KNeighborsClassifier_kd_tree	predict	1000	1000	2	0.022	0.001	0.001	0.000	-1	5	0.038	0.001	0.891	0.595	0.595	See	See
23	KNeighborsClassifier_kd_tree	predict	1000	1	2	0.002	0.000	0.000	0.002	-1	5	0.005	0.000	1.000	0.367	0.367	See	See

HistGradientBoostingClassifier_best

onnx (1.10.1) vs. scikit-learn (1.0.dev0)

Speedup barplots

All estimators share the following parameters: learning_rate=0.01, n_iter_no_change=10.0, max_leaf_nodes=100.0, max_bins=255.0, min_samples_leaf=100.0, max_iter=300.0.

WARNING! Mismatch between validation scores for 1 prediction (100.0%). See below.

	estimator	function	n_samples_train	n_samples	n_features	accuracy_score_sklearn	accuracy_score_onnx	diff_accuracy_scores
1	HistGradientBoostingClassifier_best	predict	100000	1000	100	0.824	1.0	0.176

Raw results

	estimator	function	n_samples_train	n_samples	n_features	mean_duration_sklearn	std_duration_sklearn	n_iter	iteration_throughput	latency	mean_duration_onnx	std_duration_onnx	accuracy_score_onnx	speedup	std_speedup	sklearn_profiling	onnx_profiling
0	HistGradientBoostingClassifier_best	fit	100000	100000	100	91.494	0.0	300	0.001	0.001	0.503	0.019	0.824	181.965	182.089	See	See

	estimator	function	n_samples_train	n_samples	n_features	mean_duration_sklearn	std_duration_sklearn	n_iter	iteration_throughput	latency	mean_duration_onnx	std_duration_onnx	accuracy_score_onnx	speedup	std_speedup	sklearn_profiling	onnx_profiling
1	HistGradientBoostingClassifier_best	predict	100000	1000	100	0.108	0.001	300	0.007	0.0	0.41	0.012	1.0	0.263	0.263	See	See